In this paper, they introduce RL-based DM using a novel mixture of expert language model (MoE-LM) that consists of

1. A LM capable of learning diverse semantics for conversation histories
2. A number of specialized LMs (or experts) capable of generating utterances corresponding to a particular attribute
3. RL-based DM that performs dialogue planning with the utterances generated by the experts or personality

They can have greater flexibility to generate sensible utterances with different intents and allows RL to focus on conversational-level DM.

Powerpoint for this talk

Powerpoint for this talk

Reference Paper

• Conservative q-learning for offline reinforcement learning
• Do Neural Dialog Systems Use the Conversation History Effectively? An Empirical Study